| Age | Commit message (Collapse) | Author |
|---|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14346 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Constrhash and Termhash
We need a function hash_constr (added in Term)
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14345 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
... and report changes on Term
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14344 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14330 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
We need to move eq_constr upwards
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14319 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14315 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
- removed duplicate constructors
- moved code around for more clarity
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14314 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14313 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
- moved the alterate Hashtable module to a separate file
- moved all hashconsing-related function to a separate section in Term
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14312 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14311 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
A particular case in sort-polymorphism of inductive types allows
an informative type (such as prod) to have instances in Prop:
(I,I) : True * True : Prop
This is due to the fact that prod is a singleton type: indeed (I,I)
has no informative content. But this invalidates an important invariant
for the correctness of the extraction: inductive constructors stop
having always the same sort as their inductive type. Consider for instance:
Definition f (X:Type)(x:X*X)(g:X->nat) := g (fst x).
Definition test := f _ (I,I) (fun _ => 0).
Then the inductive element (I,I) is extracted as a logical part __,
but during a strict evaluation (i.e. in Ocaml, not Haskell), this __
will be given to fst, and hence to a match, leading to a nasty result
(potentially segfault). Haskell is not affected, since fst is never
evaluated.
This patch adds a check for this situation during any Ocaml extraction,
leading for the moment to a fatal error. Some functions in inductive.ml
and retyping.ml now have an extra optional argument ?(polyprop=true)
that should stay untouched in regular Coq usage, while type-checking
done during extraction will disable this prop-polymorphism.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14256 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
The env was used for a particular case of Cbytegen.compile_constant_body,
but we can actually guess that it will answer a particular BCallias con.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14134 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
"Print Module M" prints now by default both a signature
(fields with their types) and a body (fields with their types
and transparent bodies).
"Print Module Type M" could be used both when M is a module
or a module Type, it will only display th signature of M.
The earlier minimalist behavior (printing only the field names)
could be reactivated by option "Set Short Module Printing".
For the moment, the content of internal sub-modules and sub-modtypes
are not displayed.
Note: this commit is an experiment, many sitations are still
unsupported. When such situations are encountered, Print Module
will fall back on the earlier minimalist behavior. This might
occur in particular in presence of "with" annotations, or in the
conjonction of a non-global module (i.e. functor or module type)
and internal sub-modules.
Side effects of this commit:
- a better compare function for global_reference, with no
allocations at each comparison
- Nametab.the_globrevtab is now searched according to user part only
of a kernel_name
- The printing of an inductive block is now in Printer, and rely less
on the Nametab. Instead, we use identifiers in mind_typename and
mind_consnames. Note that Print M.indu will not display anymore
the pseudo-code "Inductive M.indu ..." but rather "Inductive indu..."
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14117 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Bug-Debian: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=622882
Signed-off-by: Konstantinos Margaritis <markos@genesi-usa.com>
Signed-off-by: Stephane Glondu <steph@glondu.net>
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@14027 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
This reverts commit 33434695615806a85cec88452c93ea69ffc0e719.
Conflicts:
kernel/term_typing.ml
test-suite/success/polymorphism.v
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13998 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
- Be careful with consider_remaining_unif_problems: it might instantiate an evar, including the current goal!
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13995 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13994 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
- Add Set Typeclasses Debug/Depth n options for typeclasses eauto.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13989 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13988 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
After discussion with Bruno and Hugo, coqtop now accepts that an opaque
constant in a module type could be implemented by anything of
the right type, even if bodies differ. Said otherwise, with respect
to subtyping, an opaque constant behaves just as a parameter.
This was already the case in coqchk, and a footnote in documentation
is advertising for quite some time that:
"Opaque definitions are processed as assumptions."
Truly, it might seem awkward that "Definition x:=3" can implement
"Lemma x:nat. Proof 2. Qed." but the opacity ensures that nothing
can go wrong afterwards, since Coq is forced to ignore that the x
in signature has body "2".
Similarly, "T with Definition x := c" is now legal when T contains
an opaque x, even when this x isn't convertible with c.
By avoiding accesses to opaque bodies, we also achieve some speedup
(less delayed load of .vo final sections containing opaque terms).
Nota: the extraction will have to be adapted, since for the moment it
might access the body of opaque constants: the warning emitted when
doing that should become an error.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13987 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
when a "match" is kernel-ill-typed (probably rarely visible from end
user anyway but useful for debugging) + dead code.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13969 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13959 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
The recent experiment with -dont-load-proofs in the stdlib showed that
this options isn't fully safe: some axioms were generated (Include ?
functor application ? This is still to be fully understood).
Instead, I've implemented an idea of Yann: only load opaque proofs when
we need them. This is almost as fast as -dont-load-proofs (on the stdlib,
we're now 15% faster than before instead of 20% faster with -dont-load-proofs),
but fully compatible with Coq standard behavior.
Technically, the const_body field of Declarations.constant_body now regroup
const_body + const_opaque + const_inline in a ternary type. It is now either:
- Undef : an axiom or parameter, with an inline info
- Def : a transparent definition, with a constr_substituted
- OpaqueDef : an opaque definition, with a lazy constr_substitued
Accessing the lazy constr of an OpaqueDef might trigger the read on disk of
the final section of a .vo, where opaque proofs are located.
Some functions (body_of_constant, is_opaque, constant_has_body) emulate
the behavior of the old fields. The rest of Coq (including the checker)
has been adapted accordingly, either via direct access to the new const_body
or via these new functions. Many places look nicer now (ok, subjective notion).
There are now three options: -lazy-load-proofs (default), -force-load-proofs
(earlier semantics), -dont-load-proofs. Note that -outputstate now implies
-force-load-proofs (otherwise the marshaling fails on some delayed lazy).
On the way, I fixed what looked like a bug : a module type
(T with Definition x := c) was accepted even when x in T was opaque.
I also tried to clarify Subtyping.check_constant.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13952 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13948 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
We simply reuse the ocaml flag caml_signals_are_pending and the function
caml_process_pending_signals, and place a test at some place of the interpreter
loop (at a similar location as in ocaml byterun/interp.c).
The symbols caml_* we use are not officially made public in *.h installed
alongside ocaml, but they seem pretty stable (there since at least
ocaml 3.10, independent of arch and of byte/asm), so we access them via
"extern". For once, thanks dirty C...
In addition to that, when catching a Ctrl-C, we reset the vm via
"coq_sp = coq_stack_high" as suggested by Benjamin G.
This patch should be quite portable, it might even work in win32.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13947 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
evar, and simultaneously make type inference with universes work better.
This only exports more functions from kernel/univ, to be able to work
with a set of universe variables during type inference. Universe
constraints are gradually added during type checking, adding information
necessary e.g. to lower the level of unknown Type variables to Prop or
Set. There does not seem to be a disastrous performance hit on the
stdlib, but might have one on some contribs (hence the "Tentative").
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13905 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13896 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
be able to call term printers.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13886 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Note: I'm unsure about some subtyping error case apparently involving
aliases of inductive types (middle of Subtyping.check_inductive); I
bound it to some NotEqualInductiveAliases error, but this has to be
checked.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13885 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
that the fields had the same names but that the parameters of the
record had exactly the same names too.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13879 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13854 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13852 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
conversion.
- Fix trans_fconv* to use evars correctly.
- Normalize the goal with respect to evars before rewriting in
[rewrite], allowing to see instanciations from other subgoals.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13844 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13838 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Many of the record definitions for new safe_environment follow
the same pattern, we factorize them in a generic add_field function.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13837 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
This way, no more mixing of MBI / MPI.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13836 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
This way, no more absurd mixing of mp/kn and Prefix_equiv/Equiv to
consider, and hence no more anomaly or assert false left in Mod_subst.
As we say here, "faut pas melanger les torchons et les serviettes" ...
With these two specialized tables, efficiency might also be better.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13835 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13829 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
As said in CHANGES:
<<
The inlining done during application of functors can now be controlled
more precisely. In addition to the "!F G" syntax preventing any inlining,
we can now use a priority level to select parameters to inline :
"<30>F G" means "only inline in F the parameters whose levels are <= 30".
The level of a parameter can be fixed by "Parameter Inline(30) foo".
When levels aren't given, the default value is 100. One can also use
the flag "Set Inline Level ..." to set a level.
>>
Nota : the syntax "Parameter Inline(30) foo" is equivalent to
"Set Inline Level 30. Parameter Inline foo.",
and "Include <30>F G" is equivalent to "Set Inline Level 30. Include F G."
For instance, in ZBinary, eq is @Logic.eq and should rather be inlined,
while in BigZ, eq is (fun x y => [x]=[y]) and should rather not be inlined.
We could achieve this behavior by setting a level such as 30 to the
parameter eq, and then tweaking the current level when applying functors.
This idea of levels might be too restrictive, we'll see, but at least
the implementation of this change was quite simple. There might be
situation where parameters cannot be linearly ordered according to their
"inlinablility". For these cases, we would need to mention names to inline
or not at a functor application, and this is a bit more tricky
(and might be a pain to use if there are many names).
No documentation for the moment, since this feature is experimental
and might still evolve.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13807 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
According to B. Gregoire, this stuff is obsolete. Fine control
on when to launch the VM in conversion problems is now provided
by VMcast. We were already almost never boxing definitions anymore
in stdlib files.
"(Un)Boxed Definition foo" will now trigger a parsing error,
same with Fixpoint. The option "(Un)Set Boxed Definitions"
aren't there anymore, but tolerated (as no-ops), since unknown
options raise a warning instead of an error by default.
Some more cleaning could be done in the vm.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13806 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Before this commit, when simpl was finding the constant name
for folding some (mutual) fixpoint, this was done via some
repr_con followed by make_con. Problem: this doesn't preserve
the canonical part of a Names.constant. For instance the following
script was buggish:
Module M.
Fixpoint foo n := match n with O => O | S n => bar n end
with bar n := match n with O => O | S n => foo n end.
End M.
Module N.
Include M. (* foo, bar have here "user name" N but "canonical name" M *)
Eval simpl in (fun x => bar (S x)).
(* Anomaly: uncaught exception Failure "Cannot print a global reference". *)
(* since simpl has produce a different bar with both user and canonical N *)
TODO : check all other uses of make_con in the rest of the sources...
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13803 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
The whole standard library fits in 3 universes (instead of 22 with the
previous algorithm).
Very costy, time complexity: O(n^4) (if we assume map operations are
done in constant time), with n being the number of universe
variables. It takes ~35s for the whole stdlib.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13796 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
This is merely refactoring so that UniverseL{Map,Set} constructions
appear nicely.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13788 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13786 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13784 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
of "match" is not general enough; if there is a non dependent type
constraint, we also try w/o inversion predicate in the return clause.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13727 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
Sometimes the same (repr g u) was done in different functions
after being passed u as argument. We rather try to compute
(repr g u) as soon as possible, and then pass it instead of u.
Beware of sync issues : if g changes, arcu might become obsolete
(cf. setlt, setleq, merge ...)
Typical code around occurences of declare_univ was doing up to 3
lookups:
- is u in g ?
- if not we descend again in g to add it
- and then later repr is called on the same u.
With my safe_repr, we do one lookup if u is in g, and a lookup and
an addition otherwise. Ok, declare_univ was rarely used, but it seems
nicer this way.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13726 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
No need to tell the world about the fact that constraints are
implemented via caml's Set. Other modules just need to know about
the empty and union functions (and addition functions "enforce_geq"
and "enforce_eq" that were already there).
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13725 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
The gains on contribs are quite small, around 3% max, apart from
3 small contribs where it's about 10% (corresponding to 10s each).
With last patch, we add quicker lookup for universes in the graph
(up to 5 times less calls to cmp_univ_level on an example), but
probably more "administrative" work (i.e. addition of updated paths
in the graphs, handling pairs of updated graphs and results in
functions, etc), and some sharing might also have been lost since
graphs changed more.
Anyway, little gain and more complex code, let's remove this patch
for now ... until the next attempt to speed-up the universe layer.
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13724 85f007b7-540e-0410-9357-904b9bb8a0f7
|
|
We'll see experimentally if this helps... A few more functions could
be adapted (e.g. between), and an occurence of compare just discard
the compacted graph (in compare_greater)
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@13723 85f007b7-540e-0410-9357-904b9bb8a0f7
|
